Design and Implementation of Parallel FFT on CUDA

Fast Fourier Transform (FFT) algorithm has an important role in the image processing and scientific computing, and it´s a highly parallel divide-and-conquer algorithm. In this paper, we exploited the Compute Unified Device Architecture CUDA technology and contemporary graphics processing units (GPUs) to achieve higher performance. We focused on two aspects to optimize the ordinary FFT algorithm, multi-threaded parallelism and memory hierarchy. We also proposed parallelism optimization strategies when the data volume occurs and predicted the possible situation when the amount of data increased further.it can be seen from the results that Parallel FFT algorithm is more efficient than the ordinary FFT algorithm.